def parasites(image, cells, voronoi):
img = Functions.global_otsu(image)
cells = Functions.global_otsu(cells)
s_elem = Functions.fig(Functions.fig_size)
# Remove cells
for i in range(Functions.iterations):
cells = binary_dilation(cells, s_elem)
return_image = Functions.subtraction(img, cells)
# Remove stuff from cells
for i in range(Functions.iterations-1):
return_image = binary_erosion(return_image)
return_image = binary_opening(return_image)
for i in range(Functions.iterations - 1):
return_image = binary_dilation(return_image)
# Remove bigger objects
removal_image = return_image.copy()
for i in range(Functions.iterations + 5):
removal_image = binary_erosion(removal_image)
removal_image = binary_opening(removal_image)
for i in range(Functions.iterations + 10):
removal_image = binary_dilation(removal_image)
return_image = Functions.subtraction(return_image, removal_image)
# Remove voronoi lines for better quality
return Functions.subtraction(return_image, voronoi)
def opening3D(data, selem=skimor.disk(3), sliceId=0):
if sliceId == 0:
for i in range(data.shape[0]):
data[i,:,:] = skimor.binary_opening(data[i,:,:], selem)
elif sliceId == 2:
for i in range(data.shape[2]):
data[:,:,i] = skimor.binary_opening(data[:,:,i], selem)
return data
def edge_confidence(epi, window=9, threshold=0.02):
"""
Calculates the edge confidence according to eq. 2.
This is a simple measurement which designates an edge if pixel intensities
of gray value changes. An edge is determined by the sum of difference in
pixel intensity of a central pixel to all pixels in a 1D window of size
window. It is assumed that there is an edge if the sum is greater than a
given threshold.
Parameters
----------
epi : numpy.array [v,u]
Set of all gray-value epis for scanline s_hat.
window_size : int, optional
The 1D window siz in pixels. As the window should be centered
symmetrically around the pixel to process the value shpuld be odd. For
even numbers the next higher odd number is chosen.
threshold : float, optional
The threshold giving the smallest difference in EPI luminescence which
must be overcome to designate an edge.
Returns
-------
Ce : numpy.array [v,u]
Edge confidence values for each EPI pixel.
Me : numpy.array [v,u] of boolean.
True means an edge was discovered for at that pixel.
"""
v_dim = epi.shape[0]
u_dim = epi.shape[1]
# Check dimensions of input data
assert epi.shape == (v_dim, u_dim,), 'Input EPI has wrong shape in function \'edge_confidence\'.'
# Make window size odd
if window % 2 == 0:
warnings.warn(
'window should be an odd number in function \'edge_confidence\'. Window size {g} was given but {u} is used instead.'.format(
g=window, u=window + 1))
window += 1
# We avoid the border problem by padding the epi.
padded_epi = np.pad(epi, ((0, 0), (int(window // 2), int(window // 2))), 'edge')
assert padded_epi.shape == (v_dim, u_dim + window - 1,), 'Padded epi has wrong shape in function \'edge_confidence\'.'
# Calculate confidence values
Ce = np.zeros(epi.shape, dtype=np.float32) # initiate array
for k in range(window):
Ce += (epi[...] - padded_epi[:, k:epi.shape[1] + k]) ** 2
Me = Ce > threshold # create confidence Mask
Me = binary_opening(Me, selem=square(2), out=Me) # work with square to avoid aliasing
# Let's see if our results have reasonable meaning'
assert np.all(
Ce >= 0), 'Negative edge confidence found in function \'edge_confidence\'.'
assert Ce.shape == (v_dim,
u_dim,), 'Ce output has incorrect shape in fucntion \'edge_confidence\'.'
assert Me.shape == (v_dim,
u_dim,), 'Me output has incorrect shape in fucntion \'edge_confidence\'.'
return Ce, Me
def opening(gray_img, kernel=None):
"""Wrapper for scikit-image opening functions. Opening can remove small bright spots (i.e. salt).
Inputs:
gray_img = input image (grayscale or binary)
kernel = optional neighborhood, expressed as an array of 1s and 0s. If None, use cross-shaped structuring element.
:param gray_img: ndarray
:param kernel = ndarray
:return filtered_img: ndarray
"""
params.device += 1
# Make sure the image is binary/grayscale
if len(np.shape(gray_img)) != 2:
fatal_error("Input image must be grayscale or binary")
# If image is binary use the faster method
if len(np.unique(gray_img)) == 2:
bool_img = morphology.binary_opening(gray_img, kernel)
filtered_img = np.copy(bool_img.astype(np.uint8) * 255)
# Otherwise use method appropriate for grayscale images
else:
filtered_img = morphology.opening(gray_img, kernel)
if params.debug == 'print':
print_image(filtered_img, os.path.join(params.debug_outdir, str(params.device) + '_opening' + '.png'))
elif params.debug == 'plot':
plot_image(filtered_img, cmap='gray')
return filtered_img
def split_image_into_sudoku_pieces_adaptive_global(image, otsu_local=False, apply_gaussian=False):
L = image.shape[0]
d = int(np.ceil(L / 9))
dd = d // 5
output = []
if apply_gaussian:
image = gaussian_filter(image, sigma=1.0)
if not otsu_local:
image = to_binary_adaptive(image)
for k in range(9):
this_row = []
start_row_i = max([k * d - dd, 0])
stop_row_i = min([(k + 1) * d + dd, L])
for kk in range(9):
start_col_i = max([kk * d - dd, 0])
stop_col_i = min([(kk + 1) * d + dd, L])
i = image[start_row_i:stop_row_i, start_col_i:stop_col_i].copy()
if otsu_local:
i = to_binary_otsu(i)
i = binary_opening(i)
i = to_binary_otsu(i)
if apply_gaussian:
i = to_binary_otsu(binary_dilation(i))
this_row.append(i)
output.append(this_row)
return output, image
def outlineSmoothing(image, radius=1):
"""Smooth outlines of foreground object using morphological operations."""
struct = median(disk(radius), disk(1))
label_image = binary_closing(image, struct)
label_image = binary_opening(label_image, struct)
return label_image.astype(image.dtype)
def ProcessImage(im, targetDim = 250, doDenoiseOpening = True):
#Resize to specified pixels max edge size
scaling = 1.
if im.shape[0] > im.shape[1]:
if im.shape[0] != targetDim:
scaling = float(targetDim) / im.shape[0]
im = misc.imresize(im, (targetDim, int(round(im.shape[1] * scaling))))
else:
if im.shape[1] != targetDim:
scaling = float(targetDim) / im.shape[1]
im = misc.imresize(im, (int(round(im.shape[0] * scaling)), targetDim))
#print "scaling", scaling
greyim = 0.2126 * im[:,:,0] + 0.7152 * im[:,:,1] + 0.0722 * im[:,:,2]
#Highlight number plate
imnorm = np.array(greyim, dtype=np.uint8)
se = np.ones((3, 30), dtype=np.uint8)
opim = morph.opening(imnorm, se)
diff = greyim - opim + 128.
misc.imsave("diff.png", diff)
#Binarize image
vals = diff.copy()
vals = vals.reshape((vals.size))
meanVal = vals.mean()
stdVal = vals.std()
threshold = meanVal + stdVal
#print "Threshold", threshold
binIm = diff > threshold
misc.imsave("threshold.png", binIm)
#print vals.shape
#plt.plot(vals)
#plt.show()
#Denoise
diamond = morph.diamond(2)
if doDenoiseOpening:
currentIm = morph.binary_opening(binIm, diamond)
else:
currentIm = binIm
denoiseIm2 = morph.binary_closing(currentIm, np.ones((3, 13)))
#print "currentIm", currentIm.min(), currentIm.max(), currentIm.mean()
#print "denoiseIm2", denoiseIm2.min(), denoiseIm2.max(), currentIm.mean()
#misc.imsave("denoised1.png", currentIm * 255)
#misc.imsave("denoised2.png", denoiseIm2 * 255)
#Number candidate regions
#print "Numbering regions"
numberedRegions, maxRegionNum = morph.label(denoiseIm2, 4, 0, return_num = True)
return numberedRegions, scaling
def get_connected(landmask,TEST_CONN=True):
# find connected regions
# - land (landmask==1) is the background
# - connectivity=1, diagonal connectivity doesn't count
labels,Nlabels = msr.label(landmask,background=1,return_num=True,connectivity=1)
# ==================================================
# find largest connected region
maxlen = 0
maxlab = 0
for n in range(1,Nlabels):
N = len(labels[labels==n])
# print(n,N)
if N>maxlen:
maxlen = N
maxlab = n
print(n,N)
Landmask = 1+0*landmask
Landmask[labels==maxlab] = 0.
# ==================================================
# ==================================================
# open to avoid "singular points"
selem = np.ones((3,3))
if 1:
selem = None # default (corners=0)
Landmask = morph.binary_opening(Landmask.astype('int'),selem=selem).astype(int)
if 0:
# plot original landmask
fig1 = plt.figure()
ax1 = fig1.add_subplot(1,1,1)
I1 = ax1.imshow(landmask.transpose(),origin='lower')
fig1.colorbar(I1)
fig1.show()
if TEST_CONN:
# plot processed landmask
fig3 = plt.figure()
ax3 = fig3.add_subplot(1,1,1)
I3 = ax3.imshow(Landmask.transpose(),origin='lower')
fig3.colorbar(I3)
plt.show(fig3)
if 0:
# plot all labels found
fig2 = plt.figure()
ax2 = fig2.add_subplot(1,1,1)
I2 = ax2.imshow(labels,cmap='spectral')
fig2.colorbar(I2)
plt.show(fig2)
return Landmask
def BinMM(img,thresh,size):
#threshold at sliderval (param1)
img = thrhold(img,thresh)
structElement = morphology.rectangle(size,size)
#structElement = morphology.disk(size)
#structElement = morphology.diamond(size)
img = morphology.binary_opening(np.squeeze(img), structElement).astype(np.uint8)
return img[:,:,np.newaxis]
def get_CC_size(img):
# here a much simpler function is used as the original was way too slow
sx = ndimage.sobel(img, axis=0, mode='constant')
sy = ndimage.sobel(img, axis=1, mode='constant')
sob = np.hypot(sx, sy)
thresh = np.median(sob) * 2
sob_binary = sob > thresh
im_dilated = morphology.binary_opening(sob_binary, selem = strel(7, shape = 'disk'))
im_filled = ndimage.binary_fill_holes(im_dilated)
return np.sum(im_filled)
def noise_removal(img, radius_1=1, radius_2=2, median_iterations=3):
"""
Cleans a binary image by separating loosely connected objects, eliminating
small objects, and finally smoothing edges of the remaining objects.
The method is ``binary_opening``, ``binary_closing``, and two rounds of
``median_filter``.
Parameters
----------
img : array
a boolean ndarray
radius_1 : int
Radius of disk structuring element for opening and closing.
Default = 1, which gives 3x3 square connectivity (manhattan distance = 1).
Can also supply own boolean array.
structure_2 : int
Radius of disk structuring element for smoothing with a median filter.
Default = 2, which gives a is euclidean distance < 2*sqrt(2).
Can also supply own boolean array.
Returns
-------
A boolean ndarray of the same dimensions as ``img``.
See Also
--------
In ``scipy.ndimage``, see ``skimage.morphology.binary_opening``,
``skimage.morphology.binary_closing``, and ``skimage.filters.median``
"""
if len(np.array([radius_1]).shape) == 1:
ELEMENT_1 = _disk(radius=radius_1)
else:
ELEMENT_1 = structure_1
if len(np.array([radius_2]).shape) == 1:
ELEMENT_2 = _disk(radius=radius_2)
else:
ELEMENT_2 = structure_2
out = morphology.binary_opening(img, selem=ELEMENT_1)
out = morphology.binary_closing(out, selem=ELEMENT_1)
i = 0
while i < median_iterations:
out = filters.median(out, selem=ELEMENT_2)
i += 1
return(out)
def get_primary_object_mask(self, img, img_pk):
assert img.ndim == 3, 'Expecting 3D image, got shape {}'.format(
img.shape)
assert img_pk.dtype == np.bool
# Remove frequencies above scale of individual cells (this results in clusters near spheroid centers)
img = np.abs(
ndi.gaussian_filter(img, sigma=1 * self.factors) -
ndi.gaussian_filter(img, sigma=8 * self.factors))
img = img.max(axis=0)
img = img > filters.threshold_otsu(img)
img = img | img_pk # Merge threshold mask with given peaks/markers
img = morphology.binary_closing(img, selem=morphology.disk(8))
img = ndi.morphology.binary_fill_holes(img)
img = morphology.binary_opening(img, selem=morphology.disk(8))
# Not necessary with larger opening (min size = ~227 w/ 8 radius disk)
# img = morphology.remove_small_objects(img, min_size=256)
return img
def evaluate(image):
'''
input : RGB (png, jpg, etc.) image of neuron
output : B&W image of segmentation, B&W image of soma
'''
input = ImageOps.grayscale(image)
input = np.reshape(input, (1, 1, 512, 512))
input = torch.from_numpy(input)
input = input.type(torch.FloatTensor)
output = model(input)
output = F.sigmoid(output).data.cpu().numpy()
output[0, 1] = binary_opening(output[0, 1] > 0.8 * np.max(output[0, 1]),
disk(1)).astype(np.int)
output[0, 0] = output[0, 0] > threshold_otsu(output[0, 0])
return output[0, 0], output[0, 1]
def calculate_dice_coeff(ground_truth=Y_val, test_data=X_val, model=model):
predicted_masks = model.predict(test_data)
dice_coef_list = []
for i in range(predicted_masks.shape[0]):
pred_mask = predicted_masks[i].reshape((320, 480))
pred_mask[pred_mask >= 0.5] = 1.
pred_mask[pred_mask < 0.5] = 0.
pred_mask = ndi.binary_closing(pred_mask, np.ones((1, 1))).astype(int)
pred_mask = ndi.binary_fill_holes(pred_mask, np.ones(
(10, 10))).astype(int)
pred_mask = morphology.binary_opening(pred_mask, np.ones((10, 10)))
z = skm.confusion_matrix(
ground_truth[i].reshape((320, 480)).astype(int).flatten(),
pred_mask.astype(int).flatten())
dice_coef_list.append(2 * (z[1][1]) /
float(2 * z[1][1] + z[0][1] + z[1][0]))
print 'Avg. Dice Coeff: ' + str(np.mean(dice_coef_list))
def show_pred_mask(num):
image_rgb = misc.imread('train_resized/' + image_list[num])/255.
image_all_mask = misc.imread('train_all_angle_masks_resized/' + image_list[num].split('_')[0] + '.jpg',flatten=True)/255.
image_bw = misc.imread('train_resized/' + image_list[num],flatten=True)/255.
angle_id = int(image_list[num].split('.')[0].split('_')[1])
min_mask = misc.imread('min_and_max_masks/' + 'min_' + str(int(angle_id) - 1) + '.jpg',flatten=True)/255.
max_mask = misc.imread('min_and_max_masks/' + 'max_' + str(int(angle_id) - 1) + '.jpg',flatten=True)/255.
X = np.empty((1, 320, 480, 12), dtype=np.float32)
X[0,...,:3] = image_rgb
X[0,...,3] = image_all_mask
X[0,...,4] = min_mask
X[0,...,5] = max_mask
z = filters.sobel(image_bw)
smoothed_edges = filters.gaussian(z,1)
X[0,...,6] = smoothed_edges
z = canny(z)
z = ndi.binary_closing(z,structure=np.ones((5,5)))
X[0,...,7] = ndi.binary_fill_holes(z,np.ones((3,3))).astype(int)
threshold_otsu_val = filters.threshold_otsu(image_bw)
threshold_otsu = image_bw > threshold_otsu_val
X[0,...,8] = threshold_otsu
image_index = np.where(z >= 0)
df = pd.DataFrame()
df['l1_dist_y'] = abs((image_index[0] - 159.5)/159.5)
df['l1_dist_x'] = abs((image_index[1] - 239.5)/239.5)
df['l2_dist'] = np.sqrt((df['l1_dist_x'])**2 + (df['l1_dist_y'])**2)/np.sqrt(2)
X[0,...,9] = df.l2_dist.reshape((320,480))
X[0,...,10] = df.l1_dist_x.reshape((320,480))
X[0,...,11] = df.l1_dist_y.reshape((320,480))
pred_mask = model.predict(X).reshape((320,480))
pred_mask[pred_mask >= 0.5] = 1.
pred_mask[pred_mask < 0.5] = 0.
pred_mask = ndi.binary_closing(pred_mask,np.ones((1,1))).astype(int)
pred_mask = ndi.binary_fill_holes(pred_mask,np.ones((10,10))).astype(int)
pred_mask = morphology.binary_opening(pred_mask,np.ones((10,10)))
act_mask = misc.imread('train_masks_resized/' + image_list[num].split('.')[0] + '_mask.jpg',flatten=True)/255.
z = skm.confusion_matrix(act_mask.astype(int).flatten(),pred_mask.astype(int).flatten())
dice_coeff = 2*(z[1][1])/float(2*z[1][1] + z[0][1] + z[1][0])
print 'Dice Coeff: ' + str(dice_coeff)
fig = plt.figure(figsize = (13,13))
ax1 = fig.add_subplot(121)
io.imshow(act_mask - pred_mask)
ax2 = fig.add_subplot(221)
io.imshow(image_rgb)
plt.show()
def img_2d_stack_to_binary_array(img_2d_stack,
bth_k=3,
wth_k=3,
close_k=1,
plot_all=False,
window_size=(7, 15, 15),
slice_contrast=0,
pre_thresh='sauvola'):
print(' - Converting each z-plane to a binary mask')
img_2d_stack = np.asarray(img_2d_stack)
img_stacked = []
for im_plane in range(0, len(img_2d_stack)):
im = img_2d_stack[im_plane]
if slice_contrast:
im = st2.contrasts(im, plot=False)[slice_contrast]
im = st2.black_top_hat(im, plot=False, k=bth_k, verbose=False)
im = st2.white_top_hat(im, plot=False, k=wth_k, verbose=False)
img_stacked.append(im)
img_out = []
if pre_thresh != 'none':
thresh_3d = threshold_sauvola(np.asarray(img_stacked),
window_size=window_size,
k=0.3)
else:
thresh_3d = np.asarray(img_stacked)
for plane in range(0, len(thresh_3d)):
thresh = thresh_3d[plane] > threshold_li(thresh_3d[plane])
img_out.append(
st2.close_binary(thresh, k=close_k, plot=False, verbose=False))
img_out = np.asarray(img_out)
img_out = img_out / np.amax(img_out)
img_out = morphology.binary_opening(img_out)
img_out = morphology.binary_closing(img_out)
#img_out = st2.get_largest_connected_region(img_out, plot=False, verbose=False)
if plot_all:
tif.imsave('tmp.tiff', np.asarray(img_out, 'uint8'), bigtiff=True)
t = tif.imread('tmp.tiff')
tif.imshow(t)
plt.show()
os.remove('tmp.tiff')
return img_out
def segments_extraction(image):
"""
:param image:
:return:
"""
# бинарим изображение
binary = image < .5
# ширина и высота
w, h = image.shape
# коэффициент уменьшения изображения для создания карты сегментов
scale = 800 / max(w, h)
scaled = rescale(binary, scale)
w, h = scaled.shape
window_o = np.ones((1, int(w / 100)))
window = np.ones((8, 8))
edges = sobel(scaled)
open_image = binary_closing(edges, window_o)
close_image = binary_opening(open_image, window_o)
#
dilate = dilation(close_image, window)
#
contours = find_contours(dilate, .8)
# список для хранения сегментов
segments = []
for contour in contours:
segment = binary[int(min(contour[:, 0]) /
scale):int(max(contour[:, 0]) / scale),
int(min(contour[:, 1]) /
scale):int(max(contour[:, 1]) / scale)]
# Если в сегмента средний уровень яркости маленький, значит полезной
# информации там нет
if segment.mean() <= .05:
continue
coordinates = min(contour[:, 0]) / scale, max(contour[:, 0]) / scale, \
min(contour[:, 1]) / scale, max(contour[:, 1]) / scale
segments.append((segment, coordinates))
return segments
def create_cloud_mask(im_QA, satname, cloud_mask_issue):
"""
Creates a cloud mask using the information contained in the Quality Assessment (QA) band
for each satellite.
Arguments:
-----------
im_QA: np.array
Image containing the QA band
satname: string
short name for the satellite: 'L5', 'L7', 'L8' or 'S2'/'S2_RGB'
cloud_mask_issue: boolean
Set to True if there is an issue with the cloud mask and pixels are being
erroneously masked on the images. It will check pixels around to determine
if the cloud information is a mistake and shouldn't be taken into account.
Returns:
-----------
cloud_mask : np.array
boolean array with True if a pixel is cloudy and False otherwise
"""
# convert QA bits (the value of bits allocated to cloud cover vary depending on the satellite mission)
if satname == 'L8':
cloud_values = [2800, 2804, 2808, 2812, 6896, 6900, 6904, 6908]
elif satname == 'L7' or satname == 'L5':
cloud_values = [752, 756, 760, 764]
elif satname == 'S2' or satname == 'S2_RGB':
cloud_values = [1024, 2048] # 1024 = dense clouds, 2048 = cirrus clouds
# find which pixels have bits corresponding to cloud values
cloud_mask = np.isin(im_QA, cloud_values)
# remove cloud pixels that form very thin features. These can be for example beach or swash pixels
# that are erroneously identified as clouds by the CFMASK algorithm applied to the images by the USGS
if sum(sum(cloud_mask)) > 0 and sum(sum(~cloud_mask)) > 0:
morphology.remove_small_objects(cloud_mask, min_size=10, connectivity=1, in_place=True)
if cloud_mask_issue:
elem = morphology.square(3) # use a square of width 3 pixels
cloud_mask = morphology.binary_opening(cloud_mask,elem) # perform image opening
# remove objects with less than 25 connected pixels
morphology.remove_small_objects(cloud_mask, min_size=25, connectivity=1, in_place=True)
return cloud_mask
def thresh_and_binarize(image_,
method='rosin',
invert=True,
min_size=10,
thr_percent=95):
"""
receives greyscale np.ndarray image
inverts the intensities
thresholds on the minimum peak
converts the image into binary about that threshold
:param image_: np.ndarray
:param method: str
'bimodal' or 'unimodal'
:return: spots threshold_min on this image
"""
if invert:
image_ = u.invert(image_)
if method == 'bimodal':
thresh = threshold_minimum(image_, nbins=512)
spots = copy(image_)
spots[image_ < thresh] = 0
spots[image_ >= thresh] = 1
elif method == 'otsu':
spots = create_otsu_mask(image_, scale=1)
elif method == 'rosin':
spots = create_unimodal_mask(image_, str_elem_size=3)
elif method == 'bright_spots':
spots = image_ > np.percentile(image_, thr_percent)
str_elem = disk(min_size)
# spots = binary_closing(spots, str_elem)
spots = binary_opening(spots, str_elem)
spots = clear_border(spots)
else:
raise ModuleNotFoundError(
"not a supported method for thresh_and_binarize")
return spots
def task4():
img = mimg.imread("x.png")
img = rgb2gray(img)
img = img_as_ubyte(img)
ret, thresh = cv2.threshold(img, 150, 255, cv2.THRESH_BINARY)
new_img = binary_opening(thresh, selem=disk(35))
# fig, ax = plt.subplots(1, 3)
# show(ax[0], img, "original")
# show(ax[1], thresh, "thresh")
# show(ax[2], new_img, "new_img")
# plt.show()
fig, ax = plt.subplots(1, 1)
show(ax, new_img)
plt.show()
def post_process(img,min_size=100):
'''
图像后处理过程
包括开运算和去除过小体素
返回uint16格式numpy二值数组
'''
img = img.cpu()
img = img.numpy().astype(np.bool)
b,c,w,h = img.shape
if c==1:
for i in range(b):
img_tmp = img[i,0,:,:]
img_tmp = binary_opening(img_tmp)
remove_small_objects(img_tmp, min_size=min_size, in_place=True)
img_tmp = ~remove_small_objects(~img_tmp, min_size=min_size)
img[i,0,:,:] = img_tmp
return img.astype(np.uint16)
def get_spine_bbox(x):
s = (x > 250).sum(axis=(2))
# размытие
s_ = gaussian(s, sigma=3, preserve_range=True)
s_ = s_ / s_.max() > 0.45
# открытие морфология
s_ = binary_opening(s_, disk(4))
# наиб связ компонента
s_ = get_greatest_component(s_)
# bbox = mask2bounding_box(s_)
# s_[1][0]-=10
bbox = mask2bounding_box(s_)
bbox = add_margin(bbox, [20, 0])
b = bbox.copy()
b[0][0] -= 20
return b
def segmentTumor(path, save_path=None, return_contour=True):
name = path[path.rfind('/') + 1:path.rfind('.')]
image_array = sitk.GetArrayFromImage(sitk.ReadImage(path))
converted = convertImagePixel(image_array[0, :, :]).astype(
np.int16) # 转化为0~255的图像
# scipy.misc.toimage(converted, cmin=0, cmax=255).save(name) # 暂存灰度图
# img = cv2.imread(name)
# gray_img = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)
thresh = np.array(converted > int(255 * 0.4),
dtype=np.int8) # 变为二值图像， 可调整参数
dst_close = sm.binary_closing(thresh, sm.disk(3.5)) #闭操作滤波，填充空洞， 可调整参数
dst_open = sm.binary_opening(dst_close, sm.disk(3.5)) # 开操作滤波，消除小斑点， 可调整参数
plt.imshow(dst_open)
plt.show()
def task2():
img = mimg.imread("bw2.bmp")
# domkniecie robi najpierw dylatacje potem erozje wiec usuwa wystajace elementy
new_img = binary_closing(img, selem=disk(2.4))
# otwarcie robi najpierw erosje potem dylatacje, usuwa rysy
new_img = binary_opening(new_img, selem=disk(2))
# alternatywa
# new_img = binary_closing(img, selem=disk(2.4))
# new_img= 1-new_img
# new_img = binary_closing(img, selem=disk(2.4))
# new_img=1-new_img
fig, ax = plt.subplots(1, 2)
show(ax[0], img, "before")
show(ax[1], new_img, "after")
plt.show()
def threshold(self):
"""Threshold by first removing nuclear stain bleed through"""
dna_im = self.dreader.get_max_im(self.iminfo, 'dna')
im_subtracted = imfuns.subtract(self.im, self.C['DNA_FACTOR'] * dna_im)
im_mask = imfuns.imthresh(im_subtracted, self.C['THRESH']) > 0
im_closed = morphology.binary_closing(im_mask,
selem=morphology.disk(
self.C['MORPH_CLOSING_SZ']))
im_opened = morphology.binary_opening(im_closed,
selem=morphology.disk(
self.C['MORPH_OPENING_SZ']))
self.im_threshed = morphology.remove_small_objects(
im_opened, min_size=self.C['MIN_SZ'])
self.im_smooth = imfuns.mask_im(self.im, self.im_threshed)
return im_subtracted, im_mask, im_closed, im_opened
def bin_open(img: np.ndarray, disk_size: int) -> np.ndarray:
"""
Performs binary opening of an image
img : np.ndarray
Image to filter.
masking_radius : int
Radius of the disk-shaped structuring element.
Returns
-------
np.ndarray :
Filtered image, same shape as input
"""
selem = disk(disk_size)
res = binary_opening(img, selem)
return res
def binary_filter(image, threshold=False, percentage=0.6, size=15):
"""
:param image:
:param threshold: set threshold manually
:param percentage: get brightest percentage of image
:return: biary mask
"""
if not threshold:
ord_img = np.sort(image.flatten())
value = ceil(len(ord_img) * percentage)
if value < len(ord_img):
threshold = ord_img[value]
else:
print("threshold too high")
threshold = 0.5
binary = image > threshold
binary = binary_opening(binary, selem=disk(size))
return binary
def create_cloud_mask(im_QA, satname, cloud_mask_issue):
"""
Creates a cloud mask using the information contained in the QA band.
KV WRL 2018
Arguments:
-----------
im_QA: np.array
Image containing the QA band
satname: string
short name for the satellite (L5, L7, L8 or S2)
cloud_mask_issue: boolean
True if there is an issue with the cloud mask and sand pixels are being masked on the images
Returns:
-----------
cloud_mask : np.array
A boolean array with True if a pixel is cloudy and False otherwise
"""
# convert QA bits (the bits allocated to cloud cover vary depending on the satellite mission)
if satname == 'L8':
cloud_values = [2800, 2804, 2808, 2812, 6896, 6900, 6904, 6908]
elif satname == 'L7' or satname == 'L5' or satname == 'L4':
cloud_values = [752, 756, 760, 764]
elif satname == 'S2':
cloud_values = [1024, 2048] # 1024 = dense cloud, 2048 = cirrus clouds
# find which pixels have bits corresponding to cloud values
cloud_mask = np.isin(im_QA, cloud_values)
# remove cloud pixels that form very thin features. These are beach or swash pixels that are
# erroneously identified as clouds by the CFMASK algorithm applied to the images by the USGS.
if sum(sum(cloud_mask)) > 0 and sum(sum(~cloud_mask)) > 0:
morphology.remove_small_objects(cloud_mask, min_size=10, connectivity=1, in_place=True)
if cloud_mask_issue:
elem = morphology.square(3) # use a square of width 3 pixels
cloud_mask = morphology.binary_opening(cloud_mask,elem) # perform image opening
# remove objects with less than 25 connected pixels
morphology.remove_small_objects(cloud_mask, min_size=25, connectivity=1, in_place=True)
return cloud_mask
def segment_image(image, remove_bg=True, conv_sigma=2, opening_size=30):
"""
Segment larger regions from an image. Give it an image and it returns a
list of region objects, approximately ordered accoring the numbering at the
Excel file (i.e. from top to bottom and from left to right).
Inputs:
- image: greyscale image to segment
- remove_bg: remove background? (bool, default: True)
- conv_sigma: sigma of the Gaussian convolution (float, default: 2.0)
- opening_size: size of the patch for binary opening (int, default: 30)
Output:
- regions: the segmented regions, sorted from top to bottom and from
left to right
"""
image_greyscale = rgb2grey(image)
# smoothing
image_greyscale = gaussian_filter(image_greyscale, sigma=conv_sigma)
# erosion
seed = np.copy(image_greyscale)
seed[1:-1, 1:-1] = image_greyscale.max()
mask = image_greyscale
dilated = reconstruction(seed, mask, method='erosion')
# thresholding and binary opening
image_thresholded = dilated > 0.5 * threshold_otsu(dilated)
segmentation = binary_opening(image_thresholded,
np.ones((opening_size, opening_size)))
if 0:
distance = ndi.distance_transform_edt(segmentation)
local_maxi = peak_local_max(distance,
indices=False,
footprint=np.ones((3, 3)),
labels=segmentation)
markers = ndi.label(local_maxi)[0]
label_image = watershed(-distance, markers, mask=segmentation)
segmentation = np.logical_or(label_image > 0, segmentation)
if remove_bg:
image[segmentation == False] = 0
label_image = label(segmentation)
# find regions and sort them
regions = sort_regions(regionprops(label_image))
return regions
def closing(image, struct=None):
"""
Open up black phase so that struct fits without overlapping white phase.
:param image: data
:type image: :py:class:`numpy.ndarray`
:type struct: :py:class:`numpy.ndarray`
:type struct: :py:class:`numpy.ndarray`
:return: modified image data
:rtype: :py:class:`numpy.ndarray`
"""
if not struct:
struct = morphology.disk(1)
i = morphology.binary_opening(image, struct)
return i > i.mean()
def finalProcessingSpur(s, params):
logging.info(f"{s['filename']} - \tfinalProcessingSpur")
disk_radius = int(params.get("disk_radius ", 25))
selem = disk(disk_radius)
mask = s["img_mask_use"]
mask_opened = binary_opening(mask, selem)
mask_spur = ~mask_opened & mask
s.addToPrintList("spur_pixels", str(len(mask_spur.nonzero()[0])))
io.imsave(s["outdir"] + os.sep + s["filename"] + "_spur.png",
mask_spur * 255)
s["img_mask_use"] = mask_opened
if len(s["img_mask_use"].nonzero()
[0]) == 0: #add warning in case the final tissue is empty
logging.warning(
f"{s['filename']} - After BasicModule.finalProcessingSpur NO tissue remains detectable! Downstream modules likely to be incorrect/fail"
)
s["warnings"].append(
f"After BasicModule.finalProcessingSpur NO tissue remains detectable! Downstream modules likely to be incorrect/fail"
)
def cell_mask(Img, method):
if (method == 'Otsu'):
thresh = filters.threshold_otsu(Img)
bw = (Img < thresh)
elif (method == 'Customized'):
# rough definition of background
bw = (Img < 1.5 * np.min(Img)) # cell area
bw = ndimage.gaussian_filter(1.0 * bw, sigma=3 * DL)
bw = (bw > 0)
dc = np.mean(Img[np.where(
bw == False)]) # assume that background is a constant value
# the definition of customized extracellular boundary
thresh = division_coeff * dc + (1 - division_coeff) * np.min(Img)
bw = (Img < thresh)
bw = ndimage.binary_fill_holes(bw)
kernal = morphology.disk(np.round(DL))
bw = morphology.binary_opening(bw, kernal)
return bw
def RefineLungSegmentation(im,binary_before,binary):
binary[binary_before>0] = 0
selem = disk(5)
binary = binary_opening(binary, selem)
label_image = label(binary)
for region in regionprops(label_image):
if region.major_axis_length>60:
for coordinates in region.coords:
im[coordinates[0], coordinates[1]] = -2000
# plt.figure(13)
# plt.imshow(im, cmap=plt.cm.gray)
# plt.show()
return im
def applies_morphological_cleanup(labels,
bg_label,
morph_radius=5,
morph_min_size=25):
"""
Applies morphological cleanup to labeling assignment per component / grain.
Parameters
---------
labels : NumPy Array
labeling assignment
bg_label : int
GMM background label
morph_radius : int
disk radius for the morphological orperation
morph_min_size : int
remove small objects smaller than this value
Returns
----------
clean_labels : NumPy Array
labeling after morphological cleanup
"""
components = slu.get_unique_labels(labels)
components.remove(bg_label)
# Initialize clean labels as bacdground
clean_labels = (labels == bg_label) * bg_label
for i in components: # non-background gmm components in increasing order
gmm_labels = labels == i
clean_gmm_labels = morphology.binary_opening(
gmm_labels, morphology.disk(morph_radius))
clean_gmm_labels = morphology.remove_small_objects(
clean_gmm_labels, min_size=morph_min_size)
clean_gmm_labels = clean_gmm_labels.astype(np.uint8)
clean_labels[clean_gmm_labels == 1] = i
clean_labels[(clean_gmm_labels == 0) * gmm_labels] = bg_label
return clean_labels
def closing(image, struct=None):
"""
Open up black phase so that struct fits without overlapping white phase.
:param image: data
:type image: :py:class:`numpy.ndarray`
:type struct: :py:class:`numpy.ndarray`
:type struct: :py:class:`numpy.ndarray`
:return: modified image data
:rtype: :py:class:`numpy.ndarray`
"""
if not struct:
struct = morphology.disk(1)
i = morphology.binary_opening(image, struct)
return i > i.mean()
def threshold_segmentation(image, threshold, size=10):
"""Segment an image with a global threshold and binary opening.
Parameters
----------
image : numpy.array
Image to segment.
threshold : scalar or numpy.array
size : int
Size of a disk-shaped element to perform a binary opening.
Returns
-------
numpy.array
An array of labels.
"""
mask = image > threshold
mask = morphology.binary_opening(mask, morphology.disk(size), out=mask)
return morphology.label(mask.astype(int))
def sigmaClippedOpening(data):
"""
Perform Gaussian filtering, sigma clipping and binary opening to define
a mask for the background pixels.
"""
image = gaussian_filter(data, 0.5)
#sigma clip
masked = sigma_clip(image, sig=5., iters=None)
mask = ~masked.mask
d = masked.data
#update mask with opening
mask = binary_opening(mask, disk(8))
#plot the image
fig = plt.figure(figsize=(18, 10))
ax1 = fig.add_subplot(121)
ax2 = fig.add_subplot(122)
ax1.set_title('Data, Single Quadrant')
ax2.set_title('Background Mask')
im1 = ax1.imshow(np.log10(data), origin='lower', vmin=2., vmax=3.5, interpolation='none')
im2 = ax2.imshow(mask, origin='lower', interpolation='none')
c1 = plt.colorbar(im1, ax=ax1, orientation='horizontal')
c2 = plt.colorbar(im2, ax=ax2, orientation='horizontal', ticks=[0, 1])
c2.ax.set_xticklabels(['False', 'True'])
c1.set_label('$\log_{10}$(Counts [ADU])')
c2.set_label('Mask')
plt.savefig('opening.png')
plt.close()
out = data*mask
fileIO.writeFITS(out, 'opening.fits', int=False)
o = out.ravel()
o = o[o > 0.]
print o.min(), o.max(), o.mean(), o.std(), len(o)
return o
def get_filter_thresh(image, opening_selem=10, closing_selem=20, std_factor=1.0,
auto=False):
'''
Desc: Get image filter using thresh hold difference method.
args:
image: rgb image
opening_selem: the square matrix size for removing white speckles
closing_selem: the square matrix size for removing black speckles
auto: whether to display the image at the end of the function and prompt user for inputs
returns:
filter: black and white image with background pixels as black
'''
log.info('Getting background filter')
log.debug(f'Pixel Mean: {BG_PXL_MEAN}')
log.debug(f'Pixel STD: {BG_PXL_STD}')
# find where difference of background exceeds stddev
bg_px_map = abs(np.subtract(image, BG_PXL_MEAN)) \
<= std_factor * BG_PXL_STD
bg_px_map = np.all(bg_px_map, axis=2)
# compute filter image
filter = np.zeros(image.shape[:2])
filter[~bg_px_map] = 1
clean = binary_opening(filter, selem=square(opening_selem))
clean = binary_closing(clean, selem=square(closing_selem))
fig, (ax1, ax2, ax3) = plt.subplots(ncols=3, sharey=True, figsize = (15,10))
ax1.imshow(image)
ax1.set_title('Original')
ax2.imshow(filter, cmap='gray')
ax2.set_title('Threshold applied')
ax3.imshow(clean, cmap='gray')
ax3.set_title('Noise removed')
plt.axis('off')
fig.suptitle("Adjust threshold factor in command line to improve filter")
if not auto:
plt.show()
return clean
def get_brightest_blobs(image):
close_size = 5
open_size = 5
watershed_footprint = (5, 5)
# Make sure types are the same
input_image = img_as_float(image)
# Filter Image
filtered_image = filters.median(input_image, behavior='ndimage')
# Edge Detection
edge_sobel = filters.sobel(filtered_image)
# Threshold
thresh = filters.threshold_otsu(filtered_image)
binary_otsu = filtered_image > thresh
# Binary Morphology Operations
structure_element = morphology.disk(close_size)
closed_image = morphology.binary_closing(binary_otsu, structure_element)
structure_element = morphology.disk(open_size)
opened_image = morphology.binary_opening(closed_image, structure_element)
# Watershed
distance = distance_transform_edt(opened_image)
local_maxi = peak_local_max(distance,
indices=False,
footprint=np.ones(watershed_footprint),
labels=opened_image)
markers = label(local_maxi)[0]
labels = watershed(-distance, markers, mask=opened_image)
# Labeling and Sorting
regions = regionprops_table(labels,
intensity_image=image,
properties=('label', 'centroid',
'max_intensity', 'mean_intensity'))
df = pd.DataFrame(regions)
df = df.sort_values(by=['mean_intensity', 'max_intensity'],
ascending=False)
return df
def geometric(image_roi, attrs={}, debug=False):
""" Geometric """
image_roi, _ = pad_for_rotation(image_roi)
grayscale = rgb2gray(image_roi)
lesion_mask = image_roi[..., 3]
labels = label(lesion_mask, background=0)+1
props = regionprops(labels)
sizes = [ (p['area'], i) for i, p in enumerate(props) ]
max_value, max_index = max(sizes)
masked = np.ma.masked_array(grayscale, 1-lesion_mask)
tmin, tptp = masked.min(), masked.ptp()
# percentages reversed, lesion is darkest in the center
p25 = tptp * .75 + tmin
p50 = tptp * .50 + tmin
p75 = tptp * .25 + tmin
if debug:
print "=== Geometric ==="
plt.subplot(141)
plt.imshow(labels.copy())
for i, threshold in enumerate([p25, p50, p75], 1):
pieces = masked < threshold
pieces = binary_opening(pieces, disk(1))
labels = label(pieces, background=0)+1
N = labels.max()
attrs['Geometric p%d' % (25*i)] = N
if debug:
print "%dth percentile : %d" % (25*i, N)
plt.subplot(141 + i)
plt.imshow(labels)
if debug:
plt.show()
def geometric(image_roi, attrs={}, debug=False):
""" Geometric """
image_roi, _ = pad_for_rotation(image_roi)
grayscale = rgb2gray(image_roi)
lesion_mask = image_roi[..., 3]
labels = label(lesion_mask, background=0) + 1
props = regionprops(labels)
sizes = [(p['area'], i) for i, p in enumerate(props)]
max_value, max_index = max(sizes)
masked = np.ma.masked_array(grayscale, 1 - lesion_mask)
tmin, tptp = masked.min(), masked.ptp()
# percentages reversed, lesion is darkest in the center
p25 = tptp * .75 + tmin
p50 = tptp * .50 + tmin
p75 = tptp * .25 + tmin
if debug:
print "=== Geometric ==="
plt.subplot(141)
plt.imshow(labels.copy())
for i, threshold in enumerate([p25, p50, p75], 1):
pieces = masked < threshold
pieces = binary_opening(pieces, disk(1))
labels = label(pieces, background=0) + 1
N = labels.max()
attrs['Geometric p%d' % (25 * i)] = N
if debug:
print "%dth percentile : %d" % (25 * i, N)
plt.subplot(141 + i)
plt.imshow(labels)
if debug:
plt.show()
def grow_mask(anat, aseg, ants_segs=None, ww=7, zval=2.0, bw=4):
"""
Grow mask including pixels that have a high likelihood.
GM tissue parameters are sampled in image patches of ``ww`` size.
This is inspired on mindboggle's solution to the problem:
https://github.com/nipy/mindboggle/blob/master/mindboggle/guts/segment.py#L1660
"""
from skimage import morphology as sim
selem = sim.ball(bw)
if ants_segs is None:
ants_segs = np.zeros_like(aseg, dtype=np.uint8)
aseg[aseg == 42] = 3 # Collapse both hemispheres
gm = anat.copy()
gm[aseg != 3] = 0
refined = refine_aseg(aseg)
newrefmask = sim.binary_dilation(refined, selem) - refined
indices = np.argwhere(newrefmask > 0)
for pixel in indices:
# When ATROPOS identified the pixel as GM, set and carry on
if ants_segs[tuple(pixel)] == 2:
refined[tuple(pixel)] = 1
continue
window = gm[
pixel[0] - ww:pixel[0] + ww,
pixel[1] - ww:pixel[1] + ww,
pixel[2] - ww:pixel[2] + ww,
]
if np.any(window > 0):
mu = window[window > 0].mean()
sigma = max(window[window > 0].std(), 1.0e-5)
zstat = abs(anat[tuple(pixel)] - mu) / sigma
refined[tuple(pixel)] = int(zstat < zval)
refined = sim.binary_opening(refined, selem)
return refined
def __IsolateBody(self):
"""
create segmentation of only the human body
:return:segmentation of the human body
"""
seg = self.__activate_threshold(self.ct_mat, MIN_TH, MAX_TH)
# clean noise
seg = binary_opening(seg)
seg = remove_small_objects(seg, CLEAN_CONSTANT, connectivity=2)
# delete components until there is only one!
seg = self.__choose_largest_co_component(seg, 1)
# save nifty file
seg_file = nib.Nifti1Image(seg, self.ct_file.affine)
nib.save(
seg_file, self.dir_results + self.ct_scan + '_' + CLEAN_BODY +
'a' + NIFTY_END)
return seg
def adapt_thresh(equalized, returnmasked=False):
# Normalize by subtracting image mean
# Clip values less than zero to remove background signal
equalized = (equalized - equalized.mean()).clip(min=0)
# Otsu threshold of CLAHE equalized "grayscale"
otsu1 = threshold_otsu(equalized)
# print "Otsu threshold: {0}".format(otsu1)
thresh1 = remove_small_objects(equalized > otsu1)
colonies = thresh1*equalized
thresh2 = threshold_adaptive(colonies, 21)
# Use morphological opening to help separate clusters and remove noise
opened = binary_opening(thresh2, selem=disk(3))
if returnmasked == True:
clusters = opened*equalized
return(opened, masked)
else:
return opened
def grow_mask(anat, aseg, ants_segs=None, ww=7, zval=2.0, bw=4):
"""
Grow mask including pixels that have a high likelihood.
GM tissue parameters are sampled in image patches of ``ww`` size.
This is inspired on mindboggle's solution to the problem:
https://github.com/nipy/mindboggle/blob/master/mindboggle/guts/segment.py#L1660
"""
selem = sim.ball(bw)
if ants_segs is None:
ants_segs = np.zeros_like(aseg, dtype=np.uint8)
aseg[aseg == 42] = 3 # Collapse both hemispheres
gm = anat.copy()
gm[aseg != 3] = 0
refined = refine_aseg(aseg)
newrefmask = sim.binary_dilation(refined, selem) - refined
indices = np.argwhere(newrefmask > 0)
for pixel in indices:
# When ATROPOS identified the pixel as GM, set and carry on
if ants_segs[tuple(pixel)] == 2:
refined[tuple(pixel)] = 1
continue
window = gm[
pixel[0] - ww:pixel[0] + ww,
pixel[1] - ww:pixel[1] + ww,
pixel[2] - ww:pixel[2] + ww
]
if np.any(window > 0):
mu = window[window > 0].mean()
sigma = max(window[window > 0].std(), 1.e-5)
zstat = abs(anat[tuple(pixel)] - mu) / sigma
refined[tuple(pixel)] = int(zstat < zval)
refined = sim.binary_opening(refined, selem)
return refined
def volumeMask(vol):
"""
:param vol: a 3-dimensional numpy array
:return: mask, a binary mask with the same shape as vol, and mCoords, a list of (x,y,z) indices representing the
masked coordinates.
"""
from numpy import array, where
from scipy.signal import medfilt2d
from skimage.filter import threshold_otsu
from skimage import morphology as morph
filtVol = array([medfilt2d(x.astype('float32')) for x in vol])
thr = threshold_otsu(filtVol.ravel())
mask = filtVol > thr
strel = morph.selem.disk(3)
mask = array([morph.binary_closing(x, strel) for x in mask])
mask = array([morph.binary_opening(x, strel) for x in mask])
z, y, x = where(mask)
mCoords = zip(x, y, z)
return mask, mCoords
def _apply_threshold(img, threshold):
binary_img = img < threshold
#remove small speckles
binary_img = binary_opening(binary_img, disk(3))
return binary_img
def func(frame):
opened = binary_opening(frame, disk(smooth_size))
opened = opened & frame
return opened
from skimage import filters
import matplotlib.pyplot as plt
from scipy import ndimage as ndi
from skimage import morphology
from skimage import feature
import os
test1 = io.imread("Img09-3.tif", 1)
edge_img = io.imread("edges.png",1)
testnp = np.asarray(test1)
testnp = testnp[0:1024,0:1024]
val = filters.threshold_otsu(testnp)
mask = testnp < val
morphology.binary_opening(mask, morphology.diamond(10)).astype(np.uint8)
edges1 = feature.canny(mask)
edges2 = feature.canny(mask, sigma = 3)
fig, (ax1, ax2, ax3) = plt.subplots(nrows=1, ncols=3,figsize =(8,3), sharex = True, sharey = True)
ax1.imshow(mask, cmap=plt.cm.gray)
ax1.axis('off')
ax1.set_title('noisy image', fontsize=20)
ax2.imshow(edges1, cmap=plt.cm.gray)
ax2.axis('off')
ax2.set_title('Canny filter, $\sigma=1$', fontsize=20)
ax3.imshow(edges2, cmap=plt.cm.gray)
ax3.axis('off')
y_size = img.shape[0]
dataset = driver.Create(filename, x_size, y_size)
dataset.GetRasterBand(1).WriteArray(img)
# load original image
dataset = gdal.Open('img/mozambique-after-subset.tif')
band = dataset.GetRasterBand(1)
img = band.ReadAsArray().astype(np.uint8)
# otsu thresholding of the image
threshold = threshold_otsu(img)
img_thresholded = img > threshold
print 'Threshold for original image:', threshold
write_image(img_thresholded, 'img/mozambique-after-thresholded.tif')
# dilation of the thresholded image
img_thresholded_dilated = binary_dilation(img_thresholded)
write_image(img_thresholded_dilated, 'img/mozambique-after-thresholded-dilated.tif')
# erosion of the thresholded image
img_thresholded_eroded = binary_erosion(img_thresholded)
write_image(img_thresholded_eroded, 'img/mozambique-after-thresholded-eroded.tif')
# closing of the thresholded image
img_thresholded_closed = binary_closing(img_thresholded)
write_image(img_thresholded_closed, 'img/mozambique-after-thresholded-closed.tif')
# opening of the thresholded image
img_thresholded_opened = binary_opening(img_thresholded)
write_image(img_thresholded_opened, 'img/mozambique-after-thresholded-opened.tif')
def opening3D(data, selem=skimor.disk(3)):
for i in range(data.shape[0]):
data[i,:,:] = skimor.binary_opening(data[i,:,:], selem)
return data
num_col_strides = int_(between_fields.shape[1]/stride)
r_stride = int_(between_fields.shape[0]/num_row_strides)
c_stride = int_(between_fields.shape[1]/num_col_strides)
for r in range(num_row_strides+1):
for c in range(num_col_strides+1):
h, theta, d = hough_line(between_fields[r*r_stride:(r+1)*r_stride, c*c_stride:(c+1)*c_stride])
threshold = 0 #0.0005*max(h)
h, theta, d = hough_line_peaks(h, theta, d, min_distance=20, threshold=threshold)
for n in range(len(theta)):
if abs(theta[n]) < 0.1 or abs(theta[n]) > ((pi/2) - 0.1):
draw_hough_line(field_mask[r*r_stride:(r+1)*r_stride, c*c_stride:(c+1)*c_stride], d[n], theta[n])
# do a few small openings
field_mask = binary_opening(field_mask, rectangle(1, 5))
field_mask = binary_opening(field_mask, rectangle(5, 1))
imsave(fname_template.format('segmented_fields', 'png'), field_mask)
# Label fields
field_mask = label(field_mask, 4, 0) + 1
remove_small_objects(field_mask, 100, 1, True)
field_props = regionprops(field_mask)
# Write field_props to csv file
out_file = open(fname_template.format('field_props', 'csv'), 'w')
print('label,area,center_row,center_col,nw_col,nw_row,se_col,se_row', file=out_file)
for prop in field_props:
if (prop.area > 100) and (prop.area < 700):
print(','.join([str(x) for x in [
prop.label,
def hierarchical_segmentation(grayscale, pickle_me=False):
'''
Segments a grayscale image by first applying adaptive histogram
equalization to enhance contrast followed by an Otsu threshold to
isolate cell colonies. An adaptive thresholding method is then used to
isolate clusters of close-packed cells. Estimated regions of interest for
individual cells are finally generated using the Watershed algorithm, and
the cell regions are given unique labels and various measurements are
calculated for the regions from the original grayscale image.
Inputs:
-------
grayscale: A grayscale image loaded into a NumPy array
pickle_me: Boolean, dumps NumPy arrays of intermediate images to pickle
files in the current working directory if True
Outputs:
--------
labels: The labels associated with the thresholded regions
props: The properties of the regions-of-interest measured from the original
grayscale image
'''
# Apply CLAHE
equalized = equalize_adapthist(grayscale, ntiles_x=16, ntiles_y=16,
clip_limit=0.01, nbins=256)
# Otsu threshold of CLAHE equalized "grayscale"
otsu1 = threshold_otsu(equalized)
print "Otsu threshold: {0}".format(otsu1)
thresh1 = remove_small_objects(equalized > otsu1)
colonies = thresh1*equalized
thresh2 = threshold_adaptive(colonies, 21)
# Use morphological opening to help separate clusters and remove noise
opened = binary_opening(thresh2, selem=disk(3))
clusters = opened*equalized
# Generate labels for Watershed using local maxima of the distance
# transform as markers
distance = distance_transform_edt(opened)
local_maxi = peak_local_max(distance, min_distance=6,
indices=False, labels=opened)
markers = ndimage.label(local_maxi)[0]
# plt.imshow(markers)
# Apply Watershed
labels = watershed(-distance, markers, mask=opened)
# plt.imshow(label2rgb(labels))
# Measure labeled region properties in the illumination-corrected image
# (not the contrast stretched image)
props = regionprops(labels, intensity_image=grayscale)
# fig, axs = plt.subplots(2, 4)
# axs[0, 0].imshow(equalized, cmap=plt.cm.gray)
# axs[0, 0].set_title('CLAHE Equalized')
# axs[0, 1].imshow(thresh1, cmap=plt.cm.gray)
# axs[0, 1].set_title('Threshold 1, Otsu')
# axs[0, 2].imshow(colonies, cmap=plt.cm.gray)
# axs[0, 2].set_title('Colonies')
# axs[0, 3].imshow(thresh2, cmap=plt.cm.gray)
# axs[0, 3].set_title('Threshold 2, Adaptive')
# axs[1, 0].imshow(opened, cmap=plt.cm.gray)
# axs[1, 0].set_title('Threshold 2, Opened')
# axs[1, 1].imshow(clusters, cmap=plt.cm.gray)
# axs[1, 1].set_title('Clusters')
# axs[1, 2].imshow(distance, cmap=plt.cm.gray)
# axs[1, 2].set_title('Distance Transform')
# axs[1, 3].imshow(label2rgb(labels))
# axs[1, 3].set_title('Labelled Segmentation')
if pickle_me:
equalized.dump('CLAHE_equalized.p')
thresh1.dump('thresh1_otsu.p')
colonies.dump('colonies.p')
thresh2.dump('thresh2_adaptive.p')
opened.dump('opened_thresh2.p')
clusters.dump('clusters.p')
distance.dump('distance_transform.p')
markers.dump('max_dist_markers.p')
return(labels, props)
dest = image_source.split(image_name)[0]
if verbose:
print "\n" + pa + "*** Starting to vectorize image %s ***"%image_name
print "\n" + pa + "Current step: Image preparations"
start = time.clock() #start timer
previous_step = time.clock()
image = nh.getImage(image_source).astype(np.uint8) #Load the image.
if image.ndim > 2:
image = nh.RGBtoGray(image) #In case it is not yet grayscale, convert to grayscale.
image = np.where(image < 127,0,1) #In case it is not yet binary, perform thresholding
image = remove_small_objects(image.astype(bool),\
min_size=minimum_feature_size,connectivity=1)
if smoothing: #standard binary image noise-removal with opening followed by closing
image = binary_opening(image,disk(smoothing)) #maybe remove this processing step if depicted structures are really tiny
image = binary_closing(image,disk(smoothing))
if minimum_feature_size:
image = remove_small_objects(image.astype(bool),\
min_size=minimum_feature_size,connectivity=1)
distance_map = nh.cvDistanceMap(image).astype(np.int) #Create distance map
height,width = distance_map.shape
if save_distance_map:
scipy.misc.toimage(distance_map, cmin=0, cmax=255)\
.save(join(dest,image_name + "_dm.png")) #save the distance map in case we need it later
if debug:
scipy.misc.imsave(join(dest,image_name + "_processed.png"),image) #debug output
if verbose:
cleared__ = clear_border(nuclei__) # closed = binary_closing(cleared, disk(int(sys.argv[2]))) # # closed_ = binary_closing(cleared_, disk(int(sys.argv[2]))) # # closed__ = binary_closing(cleared__, disk(int(sys.argv[2]))) filled = ndi.binary_fill_holes(cleared) filled_ = ndi.binary_fill_holes(cleared_) filled__ = ndi.binary_fill_holes(cleared__) bopened = binary_opening(filled, disk(int(sys.argv[2]))) bopened_ = binary_opening(filled_, disk(int(sys.argv[3]))) bopened__ = binary_opening(filled__, disk(int(sys.argv[4]))) boundaries = find_boundaries(bopened, mode="inner") boundaries_ = find_boundaries(bopened_, mode="inner") boundaries__ = find_boundaries(bopened__, mode="inner") # plt.subplot(131) # # plt.imshow(hematoxylin_) #
if sigma != int(sigma):
print('skipping σ={}'.format(sigma))
continue
# get 'structureness' for each channel at each pixel
R = (K1 / K2)**2
# conservative threshold
R_thresh = ma.median(R)
targets = (R < R_thresh).all(axis=-1).filled(0)
# erode based on scale space
selem = disk(sigma)
targets = binary_opening(targets, selem=selem)
#targets = remove_small_objects(targets, min_size=selem.sum() + 1)
labeled, n_labels = label(targets, background=0, connectivity=1, return_num=True)
sizes = np.bincount(labeled.flatten())
rankings = np.argsort(sizes)
# labels for the N largest regions
N = 30 # number of regions to view
largest = rankings[-(1+N):-1]
#size_thresh = sigma**2
#largest = np.argwhere(sizes >= size_thresh).flatten()[1:]
#N= largest.size
#if N > 100:
def crop_lesion(image, attrs={}, debug=False):
""" crop image to the segmented lesion """
cleaned = binary_opening(image[...,3], disk(3))
labels = label(cleaned, 4, background=0)
props = regionprops(labels+1, intensity_image=rgb2gray(image))
sizes = [ (p['area'], i) for i, p in enumerate(props) ]
max_value, max_index = max(sizes)
p = props[max_index]
bbox = p['bbox']
image_roi = image[bbox[0]:bbox[2], bbox[1]:bbox[3], :3].copy()
alpha_roi = p['filled_image'] # or 'image'
if debug:
print """\
Image properties
----------------
Area : %d (%d filled)
Centroid : (%.3f, %.3f)
Axes : (%.3f, %.3f)
Orientation : %.3f
""" % (
p['area'], p['filled_area'],
p['centroid'][1], p['centroid'][0],
p['major_axis_length'], p['minor_axis_length'],
math.degrees(p['orientation'])
)
plt.subplot(221)
plt.imshow(image[...,3])
plt.subplot(222)
plt.imshow(labels)
plt.subplot(223)
imgplot = plt.imshow(image_roi)
imgplot.set_interpolation('nearest')
plt.subplot(224)
imgplot = plt.imshow(alpha_roi)
imgplot.set_interpolation('nearest')
plt.show()
attrs.update([
('Lesion Area', int(p['area'])),
('Lesion Area Filled', p['filled_area']),
('--Lesion Centroid X', p['centroid'][1]),
('--Lesion Centroid Y', p['centroid'][0]),
('--Lesion Orientation', math.degrees(p['orientation']) % 180),
('--Lesion Extent', p['extent']),
('Lesion Perimeter', p['perimeter']),
('Lesion Solidity', p['solidity']),
('Lesion Major Axis Length', p['major_axis_length']),
('Lesion Minor Axis Length', p['minor_axis_length']),
('--Lesion Bounding Box X1', p['bbox'][1]),
('--Lesion Bounding Box Y1', p['bbox'][0]),
('--Lesion Bounding Box X2', p['bbox'][3]),
('--Lesion Bounding Box Y2', p['bbox'][2]),
('Lesion Intensity Min', p['min_intensity']),
('Lesion Intensity Mean', p['mean_intensity']),
('Lesion Intensity Max', p['max_intensity']),
('Lesion Eccentricity', p['eccentricity']),
('Lesion FormFactor', p['area'] / (p['perimeter']**2)),
])
return np.dstack((image_roi, alpha_roi))
def predict(img, path=test_image_dir):
c_img = imread(os.path.join(path, c_img_name))
c_img = np.expand_dims(c_img, 0)/255.0
cur_seg = final_model.predict(c_img)[0]
cur_seg = binary_opening(cur_seg>0.5, np.expand_dims(disk(2), -1))
return cur_seg, c_img
# <codecell>
x = []
for i in range(3, 51, 2) :
s = filter.gaussian_filter(clean, i)
x.append(len(feature.peak_local_max(ndimage.distance_transform_edt(s > filter.threshold_otsu(s)), min_distance=5)))
plt.plot(range(3, 51, 2), x)
# <codecell>
Q = [clean]
kernelsize = 51
downsamplesize = 3
for i in range(1, 6) :
im = filter.gaussian_filter(morphology.binary_opening(Q[i-1], morphology.disk(5)), kernelsize)
im2 = zeros((im.shape[0]/downsamplesize, im.shape[1]/downsamplesize))
for x in range(im2.shape[0]) :
for y in range(im2.shape[1]) :
im2[x, y] = np.mean(im[downsamplesize*x:downsamplesize*(x+1), downsamplesize*y:downsamplesize*(y+1)])
im2 -= im2.min()
im2 /= im2.max()
Q.append(np.round(im2*1.))
# <codecell>
q = []
for i in Q :
q.append(np.sum(i) / (i.shape[0] * i.shape[1]))
plt.plot(q)